[PATCH 0/2] Allow individual features to be locked down

[Paul Moore]

On Mon, May 12, 2025 at 5:41 PM Dan Williams <dan.j.williams at intel.com> wrote:
> Dan Williams wrote:
> > Paul Moore wrote:
> > > On Fri, Mar 21, 2025 at 6:24 AM Nikolay Borisov <nik.borisov at suse.com> wrote:
> > > >
> > > > This simple change allows usecases where someone might want to  lock only specific
> > > > feature at a finer granularity than integrity/confidentiality levels allows.
> > > > The first likely user of this is the CoCo subsystem where certain features will be
> > > > disabled.
> > > >
> > > > Nikolay Borisov (2):
> > > >   lockdown: Switch implementation to using bitmap
> > > >   lockdown/kunit: Introduce kunit tests
> > >
> > > Hi Nikolay,
> > >
> > > Thanks for the patches!  With the merge window opening in a few days,
> > > it is too late to consider this for the upcoming merge window so
> > > realistically this patchset is two weeks out and I'm hopeful we'll
> > > have a dedicated Lockdown maintainer by then so I'm going to defer the
> > > ultimate decision on acceptance to them.
> >
> > The patches in this thread proposed to selectively disable /dev/mem
> > independent of all the other lockdown mitigations. That goal can be
> > achieved with more precision with this proposed patch:
> >
> > http://lore.kernel.org/67f5b75c37143_71fe2949b@dwillia2-xfh.jf.intel.com.notmuch
>
> Just wanted to circle back here and repair the damage I caused to the
> momentum of this "lockdown feature bitmap" proposal. It turns out that
> devmem maintainers are not looking to add yet more arch-specific hacks
> [1].
>
>     "Restricting /dev/mem further is a good idea, but it would be nice
>      if that could be done without adding yet another special case."
>
> security_locked_down() is already plumbed into all the places that
> confidential VMs may need to manage userspace access to confidential /
> private memory.
>
> I considered registering a new "coco-LSM" to hook
> security_locked_down(), but that immediately raises the next question of
> how does userspace discover what is currently locked_down. So just teach
> the native lockdown LSM how to be more fine-grained rather than
> complicate the situation with a new LSM.

Historically Linus has bristled at LSMs with alternative
security_locked_down() implementations/security-models, therefore I'd
probably give a nod to refining the existing Lockdown approach over a
new LSM.

Related update, there are new Lockdown maintainers coming, there is
just an issue of sorting out some email addresses first.  Hopefully
we'll see something on-list soon.

-- 
paul-moore.com